The shell game is an ancient way to get someone’s money out of their pocket and into yours. The classic example is the huckster setting up his table on the street, dockside, or at the train depot — anyplace where lots of people come by. You need a large candidate pool because you’re banking on tempting the people who combine attraction to quick riches — greed and laziness — with lower-than-average mental faculties. So if 1,000 people come by your table, you should get from 50 to 100 who would be good customers.
The game itself involves three cups (originally walnut or coconut shells probably), and a small pea or ball is placed under one. The huckster moves the shells around in ever more complex and confusing ways, relying on the customer’s likely focus on the movement of the cups.
When the huckster stops moving the cups, he then asks the customer — who has bet money that he could identify which cup covers the ball — where the ball is hidden. The customer selects. The huckster picks up the cup, revealing emptiness, and pockets the money.
The shell game relies controlling the customer’s attention to the cups, while the con man is actually concealing the fact that the ball was slid off the table on one of the slides near the table’s edge. The cups themselves have a ball attached with weak glue to the top of the cup, so that the huckster can then go to another cup, imperceptibly tap its top before lifting it, revealing the ball that has dropped to the table from its perch inside the cup.
Flash forward to the world of big data analytics.
I had the opportunity to talk to an entrepreneur who has developed a company that provides predictive analytics to describe the buying preferences of consumers related to what might be called the hedonism industry: hotels, movies, restaurants, theme parks, fashion, television and tourism.
Through the use of mobile applications that can now track location and, with user participation, human decisions related to hedonistic purchases, the idea is to provide advertisers a way to hone in on co-marketing opportunities with other entities in non-competing offerings.
For example, if you wanted to promote your movie, you could find out the connection between your type of movie and the consuming public’s choices in restaurants, clothing, airlines, television programs and so on. You’d know which TV talk show to approach to advertise your movie. You could pick which hotel chain to offer free stays at as an inducement to loyal customers of your clothing brand. Sounds good in principle, but is there a pea under the shells?
I asked the entrepreneur where they obtained the data and how they tested the validity of the correlations they concluded existed. He described Bayesian statistics, panels of experts, focus groups and databases purchased from other online sifters of human behavior data such as Facebook, Twitter, LinkedIn, Angie’s List and so on.
The more I probed into each area, the impressive claims made were apparently coming out of a sophisticated machine. But when I opened the panel on the machine, there was no machinery inside.
If you’ve seen “Men in Black II,” you know the scene where Agent J goes to find Agent K who has lost the memory of his agent life, and is a postal employee in Maine. To convince Agent K of his prior experiences, Agent J shows K that his co-workers are in fact aliens by opening up the mail sorter to find a multiarmed alien inside quickly flicking out mail through the vents to their respective distribution boxes.
That’s what I found when I opened the complex, sophisticated machinery of the big data analytics engine: lots of humans spitting out subjective judgments about relationships, the meaning of category terms and the probability of correlation.
The humans in his analytics business had to decide what categories of hedonistic purchases existed and how to define them. The humans had to decide how the buying public thought about those categories and what was meant by a “preference” as self-reported.
The subjective decisions were then built into the algorithm as surrogates for the humans, but still based on “experts” who exercised judgment. There was no validation of the predictions by tracking buying decisions of the consumers and then seeing how accurately and reliably the prediction fared.
That didn’t stop the company from presenting its results in impressive graphic form with dashboards and dials on mobile apps and computer screens, changing in real time as new “data” was fed into the software and spit back out to the various end points of users who subscribed to the service.
Those subscribers just took it on faith that when the words “algorithm,” “analytics,” “Bayesian” and “big data” were used, there was value in the tool. If the application was easy to use, pleasing to the eye, with moving colors and flashing icons … the results were taken to be real.
Marketers want to believe there’s a pea under the shell so they don’t have to spend time making tough decisions with better information. They just hand over the company’s dollar and keep their eyes on the moving shells.